Defining a comprehensive and quantitative framework for comparing urban areas is a priority for conducting research in urban ecology and related fields. In this study, we use unsupervised learning on a matrix of climate- and human-related features to estimate the number of groups explaining the characteristics of ~6,000 cities across the globe. Using estimates of city-level stability within clusters, we estimate that cities can be clustered in 5–7 clusters, with 6 being the most likely and stable number of groups. Groups of cities are primarily defined by climatic and geographical features (e.g. elevation), with population-level and land use parameters being less relevant for structuring cities. City clusters generally correspond with a continental-based partitioning. However, geographic distances between cities do not necessarily reflect their position in the examined multivariate space. The analytical framework presented in this paper can be extended to accounting for alternative features to describing cities and their characteristics. We implement an online web application for comparing cities across the globe based on the results presented in this study. This application is expected to inform decisions on where to sample populations or species in cities with either similar or divergent climatic conditions.